Image and Video Restoration and Super-Resolution
Removing blur, noise and compression damage and raising the resolution of photos and video.
Tier Grades
Tier A: Frontier AI labs
feasible
Big tech releases diffusion restorers such as SeedVR.
Tier B: Indonesian AI labs
feasible
Training GAN restorers on synthetic degradations needs no labelled data and fits an Indonesian lab.
Tier C: Institution units such as ITB
feasible
A 4-author team produced Real-ESRGAN; a university unit can work on it.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Upscaling and cleaning camera and drone images fits proofs of concept.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataNot in the sources read; Real-ESRGAN trains on purely synthetic degradations.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 4 authors on the Real-ESRGAN paper (arXiv 2107.10833).
- Time horizonYear.
- Shelf-life signalsDiffusion-based restorers replace GAN-based ones, One-step video restoration runs in real time
- Pipeline roleTier A and academia release restorers; Tier C tunes them to local cameras; Tier D uses them to improve low-quality field imagery.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
SeedVR2video generationexemplar
SwinIRimage generationexemplar
Real-ESRGANimage generationexemplar
BasicVSR++video generationexemplar
FlashVSRvideo generationproduced bySeedVRvideo generationproduced by
SUPIRvisionproduced by
DDColorimage generationproduced byCodeFormervisionproduced by
NAFNetimage generationproduced by
FILM (Frame Interpolation for Large Motion)video generationproduced byMAXIMvisionproduced byRealBasicVSRvideo generationproduced byRestormerimage generationproduced byGPENvisionproduced byGFPGANvisionproduced by
LIIFimage generationproduced byFLAVRvideo generationproduced byBasicVSR and IconVSRvideo generationproduced byGLEANimage generationproduced byRIFEvisionproduced by
TTSRimage generationproduced by
InstColorizationimage generationproduced byDIC (Deep Iterative Collaboration)image generationproduced by+25
EDVRvideo generationproduced by
TDANvideo generationproduced byESRGANimage generationproduced by
RDNimage generationproduced by
TOFlowvideo generationproduced by
EDSRimage generationproduced by
DeblurGANvisionproduced by
ESPCN (sub-pixel CNN)visionproduced bySRGAN and SRResNetimage generationproduced by
DnCNNvisionproduced by
Colorful Image Colorizationimage generationproduced bySRCNNimage generationproduced byMIRNetvisionproduced byDeOldifyvisionproduced by
QuickSRNetvisionproduced byXLSR (super-resolution)visionproduced by
Intel single-image super-resolutionvisionproduced byFBCNNvisionproduced by
SCI low-light enhancementvisionproduced by
Bringing Old Photos Back to Lifevisionproduced byFast-SRGANimage generationproduced byESRGANimage generationproduced bySeedVRvideo generationproduced byBasicVSRvideo generationproduced bySESRTinyMLproduced byLinks
R&D TopicImage and Video Restoration and Super-Resolution
ModelsReal-ESRGANTencent, image generationSwinIRETH Zurich, image generationSeedVR2ByteDance, video generationBasicVSR++Nanyang Technological University, video generation+45 more
Real-ESRGANTencent, image generationSwinIRETH Zurich, image generationSeedVR2ByteDance, video generationBasicVSR++Nanyang Technological University, video generation+45 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 49
BasicVSRNanyang Technological University, video generationproduced byESRGANThe Chinese University of Hong Kong, Nanyang Technological University, image generationproduced byESRGANThe Chinese University of Hong Kong, Nanyang Technological University, image generationproduced byRIFEMegvii, Peking University, visionproduced bySeedVRByteDance, video generationproduced bySeedVR2ByteDance, video generationexemplarTOFlowMassachusetts Institute of Technology, Google, video generationproduced byBasicVSR and IconVSRNanyang Technological University, video generationproduced byBasicVSR++Nanyang Technological University, video generationexemplarColorful Image ColorizationUC Berkeley, image generationproduced byDDColorAlibaba DAMO, image generationproduced byDnCNNHarbin Institute of Technology, The Hong Kong Polytechnic University, visionproduced by+37
EDVRThe Chinese University of Hong Kong, Nanyang Technological University, video generationproduced byFLAVRUC San Diego, Carnegie Mellon University, video generationproduced byInstColorizationNational Tsing Hua University, image generationproduced byLIIFUC San Diego, NVIDIA, image generationproduced byQuickSRNetQualcomm, visionproduced byReal-ESRGANTencent, image generationexemplarRealBasicVSRNanyang Technological University, video generationproduced byRestormerInception Institute of Artificial Intelligence, MBZUAI, image generationproduced bySeedVRByteDance, video generationproduced bySUPIRShenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shanghai AI Lab, visionproduced byTTSRShanghai Jiao Tong University, Microsoft Research, image generationproduced byBringing Old Photos Back to LifeMicrosoft AI, visionproduced byCodeFormerNanyang Technological University, visionproduced byDeblurGANUkrainian Catholic University, visionproduced byDIC (Deep Iterative Collaboration)Tsinghua University, image generationproduced byEDSRSeoul National University, image generationproduced byESPCN (sub-pixel CNN)Twitter (now X), visionproduced byFast-SRGANimage generationproduced byFBCNNETH Zurich, visionproduced byFILM (Frame Interpolation for Large Motion)Google, video generationproduced byFlashVSRTsinghua University, video generationproduced byGFPGANTencent, visionproduced byGLEANNanyang Technological University, image generationproduced byGPENAlibaba DAMO, visionproduced byIntel single-image super-resolutionIntel, visionproduced byMAXIMGoogle, visionproduced byMIRNetInception Institute of Artificial Intelligence, visionproduced byNAFNetMegvii, image generationproduced byRDNNortheastern University, image generationproduced bySCI low-light enhancementDalian University of Technology, visionproduced bySESRTinyMLproduced bySRCNNThe Chinese University of Hong Kong, image generationproduced bySRGAN and SRResNetTwitter (now X), image generationproduced bySwinIRETH Zurich, image generationexemplarTDANUniversity of Rochester, video generationproduced byXLSR (super-resolution)visionproduced byDeOldifyvisionproduced byPapers
- Image Super-Resolution Using Deep Convolutional NetworksChao Dong et al., arXiv preprint, 2014 founding
- SwinIR: Image Restoration Using Swin TransformerJingyun Liang et al., arXiv preprint, 2021 state of the art
- Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic DataXintao Wang et al., arXiv preprint, 2021 state of the art
- Restormer: Efficient Transformer for High-Resolution Image RestorationSyed Waqas Zamir et al., CVPR 2022 (per arXiv comment), 2021 state of the art
- ESRGAN: Enhanced Super-Resolution Generative Adversarial NetworksXintao Wang et al., ECCV 2018 (per arXiv comment), 2018 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2107.10833 https://arxiv.org/abs/2107.10833
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Image and Video Restoration and Super-Resolution", https://aurion.semai.id/topics/image-and-video-restoration/, records of October 2026.